Localização de Faltas em Linhas de Transmissão Baseado em Processamento Estatístico de Sinais e Aprendizado de Máquina

Abstract

This work presents a method for locating faults in Transmission Lines based on Independent Component Analysis (ICA) with an emphasis on the Second Order Blind Identification (SOBI) algorithm. The database consists of 7920 three-phase voltage signals for single-phase faults simulated via the Alternative Transient Program (ATP), at a sampling rate of 200 kHz, with different configurations of resistance, incidence angle, location and also with noise insertion, on a transmission line with a length of 200 km. The signals were segmented and processed by SOBI to be presented to the K-Nearest Neighbors (KNN) regression algorithm to estimate the fault location. Two approaches were proposed: the first uses the separation matrix obtained by fixed SOBI and the second applies SOBI to each processed signal window, generating an adaptive separation matrix. The results showed an average relative error of less than 1 km for some situations with the presence of noise.